Key Takeaways

• Unreleased animations extracted from Google App v17.23.33 reveal that “Gemini Spark” and “Gemini Agent” have two distinct visual identities: a diamond icon for Spark and a mouse pointer icon for Agent.
• Agent’s setup flow contains an onboarding screen warning about remote browser data and remote code execution. Spark has no equivalent.
• Even Google AI Ultra subscribers (at $250 per month) will likely face hard usage caps for Spark and Agent usage.

Unreleased animations extracted from the Android Google App version 17.23.33 show that the company is launching two distinct AI agents, Gemini Spark and Gemini Agent as part of Android’s recently-announced Gemini Intelligence agentic AI suite. They’re visually distinct product tiers with different icons, different onboarding flows (the initial setup screens where users agree to permissions and see how the tool works).

In my previous report I talked about the skill system and task scheduler that Gemini Intelligence runs behind the scenes. Now, in version 17.23.33 of the Google App, new user-facing front end elements have been added that weren’t present in the previous version, indicating active preparation ahead of today’s big Google I/O reveal.

Gemini Intelligence: Two Icons, Two Tools

Google has created two separate “Lottie” animation files, one for each agent:

• assistant_robin_agent_spark_working_blue.json — incorporating Gemini Spark’s four-pointed star icon
• assistant_robin_agent_working_blue.json — showing a mouse pointer icon

This points to two distinct tools with different visual identities, meaning they look very different while the user is waiting for a task to complete. The mouse pointer icon is a strong indication that the agent will be able to control the user’s device autonomously by clicking on on-screen elements.

Here’s a screenshot and short video clip of the two icons in action.

The setup process presents users with a series of welcome screens known as onboarding cards. The first two cards are identical for both Spark and Agent, providing information about data sources and the much-reported warning about automated online purchases.

However, the Agent setup presents an additional third card warning users that it will save “remote browser data, like login details and remote code execution data” to perform tasks. Gemini Spark has no such warning screen, strongly suggesting that only Agent gains the power to browse and run code on your behalf.

The Agent’s third onboarding card text reads as follows:

"To help you get things done efficiently, Gemini saves remote browser data, like login details and remote code execution data."

Gemini Ultra Users Won’t Get Unlimited Usage

Agent usage will come with limited usage quotas even for subscribers to Google’s $250 per month Google AI Ultra tier aimed at power users.

A quota banner reveals the following text regarding token limits.

For the Ultra tier:

• assistant_agent_quota_banner_body_ultra - “Gemini Spark will be available again when your limit resets.

For the free tier:

• assistant_agent_quota_banner_body_basic - “Some features aren't available until your limit resets. Upgrade for higher limits and more.”

This text strongly suggests that Gemini Intelligence will run subject to limited token quotas according to the user’s subscription plan. These will refresh periodically in a similar fashion to how Google Antigravity makes users wait for a pre-determined time if they exhaust the token allocation associated with their subscription tier.

This means even power-users spending $250 per month for Google’s most expensive Ultra tier will find themselves eventually locked out of some agentic features if their usage is heavy enough. I couldn’t find any reference of a middle-ground “Pro” tier in the code, a hint that some agentic tools may be reserved for Ultra subscribers only.

I couldn’t find anything in the code to indicate that users will be able to top up their credits on a pay-as-you-go basis as they can with Antigravity. This positions the feature as a consumer function rather than a business tool that can be relied upon to work at all times. The actual quotas aren’t hard-coded into the app, so Google will be able to adjust them from the server side on demand. We’ll have to wait for the official product launch to see the actual quotas.

What To Expect At Google I/O 2026

Google I/O 2026 is just a few hours away and the company has today already started pushing out updated user experiences for Gemini, both on mobile and desktop. Expect to see a raft of features under the Gemini Intelligence umbrella across all platforms. Mac users won’t be left out either, as Testing Catalog recently revealed in an early look at the next Gemini desktop upgrade.

The Google app also contains hidden home-screen showcase cards for AI video and image generation, pointing to a launch of new generative AI features later today.